Multiclass Classification of Alzheimer’s Disease Using Hybrid Deep Convolutional Neural Network

Abstract

Deep learning has gained a lot of attention in recent years for solving problems in several fields, including medical image processing.  This paper proposes a very deep pipeline that creates a convolutional neural network-based pipeline using magnetic resonance imaging (MRI) scans to detect Alzheimer’s disease and its stages. Diagnosing Alzheimer's disease in the elderly is difficult due to identical brain structures and pixel strength, and it necessitates systematic discrimination. Alzheimer's disease is a progressive loss of morality that results in issues such as a gradual deterioration in thought, actions, and social skills, making it difficult for an individual to work independently. Based on input images of Magnetic Resonance Imaging (MRI), we have developed an in-depth study model for predicting individual diagnosis of Alzheimer's Disease (AD), Cognitive Normal (CN) and Mild Cognitive Impairment (MCI). We have used four classes in this paper: Non-Demented, Mild Demented, Very Mild Demented, and Moderate Demented. To diagnose Alzheimer's disease, we used a combination of Inception and ResNet formulation. This proposed model achieves an accuracy of 79.12 percent, which is a significant improvement over the existing model.